SDK examples

The Python and TypeScript SDKs expose the same desktop contract. In TypeScript the desktop methods also live on computer.desktop. Method names are case-sensitive.

Method reference

Group	Python	TypeScript	Purpose
Screen	`screenshot()`	`screenshot()`	Full desktop as PNG bytes
	`screenshot_base64()`	`screenshotBase64()`	Same, base64, ready for a vision model
	`screenshot_region(x, y, w, h)`	`screenshotRegion(x, y, w, h)`	Region as PNG bytes
Click	`click(x, y, button="left")`	`click(x, y, button)`	Click at pixels; button `left`\|`right`\|`middle`
	`left_click(x, y)`	`leftClick(x, y)`	Left-button click
	`right_click(x, y)`	`rightClick(x, y)`	Right-button click
	-	`middleClick(x, y)`	Middle-button click
	`double_click(x, y)`	`doubleClick(x, y)`	Double-click
Mouse	`move_cursor(x, y)`	`moveMouse(x, y)`	Move the pointer without clicking
	`mouse_down(x, y, button="left")`	`mouseDown(x, y, button)`	Press and hold a button
	`mouse_up(x, y, button="left")`	`mouseUp(x, y, button)`	Release a held button
	`drag(from_x, from_y, to_x, to_y)`	`drag(fromX, fromY, toX, toY)`	Click-drag between two points
Keyboard	`type(text)`	`type(text)`	Type literal text at the focus
	`key(key)`	`key(key)`	Press one key or a `+` chord
	`hotkey(*keys)`	`hotkey(...keys)`	Simultaneous combo
	`key_down(key)` / `key_up(key)`	`keyDown(key)` / `keyUp(key)`	Hold or release a key
Scroll	`scroll(direction, clicks)`	`scroll({ direction, clicks })`	Scroll `up`\|`down`\|`left`\|`right` by N notches
	`scroll_up/down/left/right(clicks)`	-	Convenience scrolls
Clipboard	`get_clipboard()` / `set_clipboard(text)`	`getClipboard()` / `setClipboard(text)`	Read or write clipboard text
Screen info	`get_screen_size()`	`screenSize()`	Resolution `{ width, height }`
	`get_cursor_position()`	`cursor()`	Cursor `{ x, y }` in screen pixels
Windows	`windows()`	`windows()`	List open windows
	`launch(app)`	`launch(appName)`	Open an installed app
	`focus_window(id)`	`focusWindow(id)`	Bring a window to the front
	`get_window_size(id)` / `set_window_size(id, w, h)`	`windowSize(id)` / `resizeWindow(id, w, h)`	Read or set window size
	`get_window_position(id)` / `set_window_position(id, x, y)`	`windowPosition(id)` / `moveWindow(id, x, y)`	Read or set window position
	`maximize_window(id)` / `minimize_window(id)` / `close_window(id)`	`maximizeWindow(id)` / `minimizeWindow(id)` / `closeWindow(id)`	Change window state
Environment	`get_desktop_environment()`	`environment()`	DE name and version
	`set_wallpaper(path)`	`setWallpaper(path)`	Set the background from a VM file
	`get_accessibility_tree()`	`accessibilityTree()`	AT-SPI element tree
Shell and files	`bash(cmd)` / `python(code)`	`bash(cmd)` / `python(code)`	Run inside the VM
	`write_file(path, content)` / `read_file(path)`	`writeFile(path, content)` / `readFile(path)`	Read or write a VM file

First loop

Feed a screenshot to a vision model

Screenshot_base64 returns a string you can pass straight through as image data.

b64 = computer.screenshot_base64()

import anthropic
client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-opus-4-5",
    max_tokens=1024,
    messages=[{
        "role": "user",
        "content": [
            {"type": "image", "source": {"type": "base64", "media_type": "image/png", "data": b64}},
            {"type": "text", "text": "What is on screen, and where should I click?"},
        ],
    }],
)

Keyboard and mouse

Clipboard, scroll, and windows

Shell, Python, and files

These run inside the VM with no GUI interaction and are usually the fastest way to do multi-step work.

print(computer.bash("ls /workspace/Desktop"))
print(computer.python("print(1 + 1)"))

computer.write_file("/workspace/script.py", "print('hello')")
computer.bash("python3 /workspace/script.py")
print(computer.read_file("/workspace/.bashrc"))

CLI

miosa api GET "/computers/$COMPUTER_ID/desktop/screenshot" > before.png
miosa api GET "/computers/$COMPUTER_ID/desktop/accessibility-tree"
miosa api POST "/computers/$COMPUTER_ID/desktop/click" -d '{"x": 500, "y": 500}'
miosa api GET "/computers/$COMPUTER_ID/desktop/screenshot" > after.png